Force field is widely used to model the potential energy in atomistic simulation systems. Despite force fields have a concise mathematical form, a good set of force field parameters usually requires extra care of calibration. Besides, numerous ionic force field parameters are reported from various sources as researchers have specific target properties for their interests. Previous studies mainly used brute force optimization to find the most desired set of parameters in ionic solution. However, these methods are not efficient since the evaluation of the performance of a parameter set is time-consuming. This work used a stochastic optimization routine in machine learning to tackle the problem of black-box function optimization. This method shows excellent performance of locating the optimum regions of the black-box cost function in only a few iterations. To evaluate the performance of a set of ionic force field parameters, MD simulations are carried out in LAMMPS to compute ionic properties. The solvation free energy and ion oxygen distance are selected as the primary targets while the self-diffusion coefficient and contact ion pairs are regarded as the secondary targets. The optimum region of primary targets are found by direct optimization, then secondary targets are studied with optimized parameters of the primary targets. There have been found discrepancies between the optimum regions of different targeted properties. The dependence studies of individual ionic force field parameters ($\epsilon, \sigma, q$) are analyzed and parameterization trends are found out. Base on these trends, the final calibration model is proposed.